A birthday clip wants a voice. A caption on screen says the words, but nobody hears them, and for something sent to a family group that difference is most of the warmth. The editor can now generate that voice, which it could not do the last time this series looked.
This piece builds the clip at 16:9 — one generated still, one five-second animation, one spoken greeting — and then measures the thing that decides how the greeting is written: how many words actually fit.
The finished clip: 1280 by 720, 30 frames per second, 5.07 seconds, with a greeting spoken by a library voice. Shown here as frames in sequence; the full video and a GIF version are supplied alongside this document. The character, the cake and the room were generated for this article.
The Character, Made Once
The character is one generated picture, and it is worth being plain about what it is: an invented toy, not a licensed one. The trend this page comes from is built around a branded character, and reproducing that character's shape, coloring or face is the one thing to avoid. Use an original toy character with its own shape, color palette, facial features, and name. Do not reference, reproduce, or closely imitate recognizable branded characters, logos, packaging, or signature design traits.
A rounded 3D toy character stands on a pale wooden table, facing the camera. It is about as tall as it is wide, made of soft matte mint-green vinyl, with two small round ears on top, two large round black eyes, a tiny stitched smile and short stubby arms. It wears a small red cone party hat with a white pompom. In front of it on the table sits a small round cake with one unlit candle. The wall behind is plain warm cream, softly out of focus. Even soft light from the front, no hard shadows, matte surfaces, no glare. Sharp focus on the toy, photographed at its eye level. No logos, no printed words and no packaging anywhere in the picture. The image fills the entire frame edge to edge, no white border.
The character, generated once at 16:9 and 2K. Nothing about its shape or coloring is taken from an existing toy.
That picture goes back in as the first frame of a five-second animation: arms up, one bounce, a lean toward the cake.
Where the Voice Comes From
Add a text layer in the editor and the panel on the right gains a tab called Text to speech, sitting between Basic and AI avatars. It reads the text already in the layer — there is no second box to type into — and turning it into speech takes one click on a voice name. Select the text layer, open Text to speech, choose an available voice, and follow the generation control shown in your current editor. Preview the result and check its duration on the timeline before final export.
The speech arrives as its own clip on the audio track a few seconds later, and the caption stays where it was. That is the whole operation.
Most of the Voice Categories Are Empty
The panel opens on a Category dropdown set to Trending, and Trending is empty — it shows a placeholder reading "There's nothing yet". So does English. That is the state this series found the last time it looked, and it is easy to conclude from those two that the feature does not work.
The category dropdown decides whether there is anything to pick. Trending is the default.
Voice categories, available voice names, labels, and usage options can vary by account, language, region, and product version. Choose a voice available in your current workspace, generate a short sample, and check pronunciation, pacing, timeline length, and any displayed usage conditions before using it in a public or commercial video. For commercial distribution, review the applicable CapCut terms and licensing information and the rules of the platform where you plan to publish.
Eleven Words Do Not Fit a Five-Second Shot
The greeting written for this clip was "Happy birthday, Nora. Make a wish and blow out the candle." Eleven words, which reads as a short line. Spoken, it is not short.
The project total before the speech was 5.03 seconds, the length of the video clip. Generating speech longer than that stretches the timeline.
The timeline is the readout here. Before the speech, the project runs 5.03 seconds, because that is how long the animation is. After generating the eleven-word greeting it ran 5.43 seconds — the speech is longer than the picture, and the extra four tenths of a second play out over nothing. Cutting the line back to six words, "Happy birthday, Nora. Make a wish.", left the total at 5.03 seconds: the speech fits inside the shot.
Nothing in the panel warns about this. The number at the top of the timeline is the only place it shows up, and it is easy to miss because the change is small.
The Same Text Twice Gives Two Different Lengths
The eleven-word line was generated twice, from the same text and the same voice. The first run stretched the timeline to 5.70 seconds and the second to 5.43 seconds — a difference of nearly three tenths of a second between two runs of the same sentence.
That matters for planning more than it looks. If you write a greeting that ends exactly on the last frame, a regenerated take can push past it. Leaving half a second of margin costs nothing and removes the problem.
What the Two Requests Cost
The cost lines as shown under each request, read on 14 September 2026.
Twenty-six credits for the footage. The editor does not show a cost line of its own, so the speech, the caption and the export are not priced on screen the way a generation request is; No separate price appeared for the speech workflow in this session. That display is not a guarantee that every voice, account, region, or future session has the same availability or price; review the current label and any applicable terms before publishing.. These figures are for these configurations on this account and this date and come from the line the composer shows before each request; other models, durations, sizes and modes are charged differently, and prices change.
The still and the clip were made through the AI image generator and AI video generator flows, and the greeting was assembled in the web editor.
How Much Birthday Greeting Fits
Six words fitted a five-second shot and eleven did not, so the working figure is about one word per half second with a little room at each end. Write the line first, count it, and size the shot to the line rather than the other way round — a greeting cut off mid-word is worse than a clip that runs a second long.
If the line has to be longer, the cheaper fix is a second shot rather than a faster voice. Another five-second clip from the same still costs one more video request and gives the greeting somewhere to land; squeezing eleven words into five seconds gives you a rushed reading and a caption nobody can follow.
The planning row behind this page describes a child in the clip: a toy character celebrating alongside the child whose birthday it is, built from a photo of that child. There is no child here, generated or otherwise, and no photograph of a real person went into any of it. If you make one of these for your own family, the guidance in that row is worth following — use only your own child's photo, get the other parent's agreement before posting, and prefer the family group over a public feed.
Also not covered: automatic captions and caption templates, the AI avatars tab that sits next to Text to speech, custom voice cloning, beat-matched cutting, transitions between shots, and 4K export.
Written 14 September 2026. The character picture and the animation were generated in CapCut for this article and the greeting was assembled in its web editor; the toy is invented and no real person, child or branded character appears. Interface labels, voice lists and cost lines reflect a single session on that date and the configurations named above, and may change. Speech lengths are read from the project total shown at the top of the timeline before and after the speech clip was generated, against a video clip of 5.03 seconds; the generated clip carries its own audio track, so the spoken words were not separated from it by measurement.